Re: RC4 on AMD64

From: Jose Castejon-Amenedo (Jose.Castejon-Amenedo_at_hp.com)
Date: 11/04/04

  • Next message: Jubilation T. Cornpone, CSA: "Re: Printers may leak the printed information through side-channels: a verifiable electronic voting glitch ?" 
    Date: Thu, 04 Nov 2004 16:05:17 GMT

    On Thu, 04 Nov 2004 10:55:51 +0100, Arnaud Carré wrote:

    >> Which is not what happens with the IA64 code I mentioned. The performance
    >> quoted is sustained performance, on any input data.
    >
    > Do I read the same article ? The link you mention specify (
    > http://etud.epita.fr/~bevand_m/papers/rc4-amd64.html ):
    >
    > rc4-amd64 (rc4speed, provided with rc4-amd64) Processor Throughput (MB/s)
    > Opteron 244 1.8 GHz (64-bit) 319
    >
    >
    >
    > So the speed test giving "319Mb/s" is "rc4speed.c" as mentioned. Just open
    > "rc4speed.c" and you can see the inner loop test:
    >
    > #define BUFSIZE ((long)1024)
    > static unsigned char buf[BUFSIZE];
    >
    > // Speed counting
    > Time_F(START);
    > for (count=0,run=1; COND(cc); count++)
    > RC4(&sch,BUFSIZE,buf,buf);
    > d=Time_F(STOP);
    >
    > Timing is done on repeating RC4 on a 1024 bytes byffer len (of course the
    > test does it several times, but always in the same small buffer). So in that
    > case, primary CPU cache is the key of the speed !
    >
    > Am I wrong somewhere ??
    >
    > PS: the article does not say how "OpenSSL" speed was measured. I guess not
    > with such a small buffer :-))

            Nope, we are saying the same thing. My comment meant that, unlike the
    code for AMD, which comes with strings attached, as you point out, the
    IA64 I am referring to does not have such constraints.

  • Next message: Jubilation T. Cornpone, CSA: "Re: Printers may leak the printed information through side-channels: a verifiable electronic voting glitch ?"

    Relevant Pages

    • Re: RC4 on AMD64
      ... > quoted is sustained performance, on any input data. ... "rc4speed.c" and you can see the inner loop test: ... but always in the same small buffer). ...
      (sci.crypt)
    • Re: Simulink-buffer
      ... not come out of the buffer until the 2nd frame period). ... what if i have 100 input data and the buffer ... Zeros to the following frame?? ...
      (comp.soft-sys.matlab)
    • Re: what does acmStreamSize do?
      ... have enough input data to produce any of output data. ... Then you use the same buffer for the output of one ... > conversion and for the input of the next. ...
      (microsoft.public.win32.programmer.mmedia)
    • Question about CryptEncrypt and CryptDecrype API functions
      ... Then the Buffer to copy the input data is allocated at dwBlockLen + ... I understand that each block of data that will be copied into the buffer ... The encryption does not depend on the key length and algo used (I use 3DES ...
      (microsoft.public.security)
    • Question about CryptEncrypt and CryptDecrype API functions
      ... Then the Buffer to copy the input data is allocated at dwBlockLen + ... I understand that each block of data that will be copied into the buffer ... The encryption does not depend on the key length and algo used (I use 3DES ...
      (microsoft.public.win2000.security)